Composition-Aware Cropping and 360° View Framing with Explainable Photographic Assessments

Select a direction, refine a crop, and explain the photographic decision with measured visual attributes.

1MediaTek Inc. 2National Taiwan University 3National Yang Ming Chiao Tung University

Signal Hill sunrise

Pan +165°Tilt −0°FOV 60°
S63C79A37I25

Strongest: Composition. Weakest: Integrity. The composition excels with its balanced arrangement and adherence to the rule of thirds, placing the car centrally while allowing for a dynamic leading line that guides the viewer's eye towards the mountainous backdrop. Despite the subject being small within the frame, its placement ensures it remains prominent against the vast landscape. Integrity is compromised due to the slight zoom-out, which cuts off some of the surrounding environment but does not detract significantly from the overall composition's effectiveness.

Spruit sunrise

Pan −45°Tilt +0°FOV 60°
S54C59A48I16

Strongest: Composition. Weakest: Integrity. The composition is solid, with a balanced arrangement that uses leading lines effectively to guide the viewer's eye through the scene. The subject, a prominent tree, is well-placed within the frame, drawing attention without being overwhelming. The overall balance and use of space create a pleasing visual flow, enhancing the aesthetic appeal despite the low integrity score.

Overview

One panorama, one reproducible photograph

PhotoGuider turns a full 360° environment into a selected viewpoint, a refined crop, camera parameters, and an explanation grounded in measured photographic attributes.

PhotoGuider in practice. Scan the panorama, select a viewing direction, refine the framing, and return a reproducible photograph with an interpretable assessment.
120candidate views24 pan positions × 5 tilt positions
60°working viewComposition is refined after direction selection
3camera parameterspan, tilt, and field of view
4assessment tokensSubject, Composition, Aesthetics, Integrity

Method

From viewing direction to explained framing

The system separates where to look from how to frame, then exposes the photographic evidence behind its final decision.

Method overview. Six vision models produce four photographic tokens that guide view generation, ranking, refinement, and the final explanation.
S

Subject

How clearly the main visual subject is identified and emphasized.

C

Composition

Balance, placement, symmetry, leading lines, and spatial organization.

A

Aesthetics

The visual appeal of the selected view as a complete photograph.

I

Integrity

Whether important content remains complete and avoids unintended truncation.

2D cropping benchmarks

Competitive cropping performance

Comparison with state-of-the-art methods on FCDB and FLMS datasets.
MethodPublicationFCDBFLMSAverage
BDE↓IoU↑BDE↓IoU↑BDE↓IoU↑
DICICCV 20170.0900.6300.0570.8100.0740.720
DIC*TPAMI 20190.0800.6500.0520.8300.0660.740
A2RLCVPR 20180.0890.6640.0450.8210.0670.743
A3RLTIP 20190.0770.696-0.839-0.768
VPNCVPR 20180.0730.7110.0440.8360.0590.774
CACNetCVPR 20210.0740.7020.0370.8410.0560.772
GAICTPAMI 20220.0810.6740.0410.8340.0610.754
CAGRTOMM 20250.0670.7210.0330.8510.0500.786
InstructCropMM 20250.0700.7160.0330.8470.0520.782
CropperCVPR 20250.0620.7560.0410.8210.0520.789
VenusCVPR 20260.0710.7130.0300.8700.0510.792
CPICarXiv 20260.0580.7510.0370.8420.0480.797
PhotoGuider (fast)20260.0580.7590.0410.8350.0500.797
PhotoGuider (full)20260.0560.7660.0330.8560.0440.811
Orange PhotoGuider crop Green ground truth

User study on composition explanations

50 participants compared explanations from three methods across five scenes, making 250 choices. All methods explained the same PhotoGuider full-mode crop, isolating explanation preference from crop quality.

40.0%PhotoGuider100 / 250 choices
32.8%Venus82 / 250 choices
27.2%ArtiMuse68 / 250 choices

PhotoGuider received the most votes in all five scenes. Method names were hidden and presentation order was randomized for each participant.

Original image with the PhotoGuider crop boundary
Original image with crop boundary
PhotoGuider crop explained by all three methods
Shared crop for all three explanations

PhotoGuider

20 / 50 votes

The subject sits in the left third, about 35% across the frame, which adheres to the rule of thirds for balanced composition. The subject occupies a moderate share of the frame, filling about 17%, ensuring it remains prominent without overwhelming the image. The framing is kept as-is, maintaining all necessary elements within the frame without any significant cuts.

Venus

16 / 50 votes

The crop in the image is well-composed due to its focus on the subject, creating a strong focal point. The use of leading lines from the ceiling and the vertical lines of the light beams draw the viewer's eye directly to the subject, enhancing the sense of depth and movement. The red dress contrasts sharply with the dark background, adding visual interest and making the subject stand out. This composition effectively uses the rule of thirds, placing the subject off-center for a more dynamic and engaging image.

ArtiMuse

14 / 50 votes

The image exhibits a strong composition with a clear focal point on the subject, who is positioned slightly off-center, creating a dynamic asymmetry. The use of lighting, with the bright overhead lights contrasting against the darker background, adds depth and highlights the subject's features. The red polka-dotted top stands out vividly against the muted surroundings, drawing attention to the individual's expression and posture. The background elements, such as the ceiling and other indistinct figures, provide context without distracting from the main subject. The overall balance and contrast effectively convey a sense of introspection and resilience, aligning with the narrative of overcoming adversity.

Explanations are reproduced verbatim from the study. Each participant selected one explanation per scene.

User study on 360° framing

People preferred PhotoGuider across diverse scenes

Participants ranked crops from PhotoGuider, Venus, and CACNet without seeing method labels.

120participants
4,820rankings
100scenes
59scenes won
MethodRank 1Rank 3Mean rankScenes won
PhotoGuider53.2%15.2%1.6259
Venus27.4%37.4%2.1026
CACNet19.5%47.4%2.2815

65.2%PhotoGuider over Venus

72.8%PhotoGuider over CACNet

p = 3.7 × 10-10Friedman scene-level test

Top 10 preferred scenes

Complete decisions, not isolated crops

Each case keeps the video, competing crops, photographic assessment, camera parameters, explanation, and 120-view preference maps together.

Rank 1

Royal esplanade

95.8% ranked PhotoGuider first 48 rankings

PhotoGuider crop for Royal esplanade
PhotoGuider95.8% rank 1pan 150.4° / tilt -0.2° / fov 59.5°
Venus crop for Royal esplanade
Venus0.0% rank 1pan 79.5° / tilt -29.9° / fov 53.8°
CACNet crop for Royal esplanade
CACNet4.2% rank 1pan 179.9° / tilt -30.8° / fov 58.5°
S62C57A37I20

Strongest: Subject. Weakest: Integrity. The subject, the interior of a shopping mall, is prominently featured, drawing attention to its architectural details and layout. The composition is balanced, with a clear view of the main atrium and surrounding shops, creating a sense of depth and openness. However, the integrity score is low due to the lack of complete scenes or elements within the frame, which might be partially obscured or incomplete.

PhotoGuider view preference heatmap for Royal esplanade
PhotoGuider
Venus view preference heatmap for Royal esplanade
Venus
CACNet view preference heatmap for Royal esplanade
CACNet
Rank 2

Golden bay

94.2% ranked PhotoGuider first 52 rankings

PhotoGuider crop for Golden bay
PhotoGuider94.2% rank 1pan 105.4° / tilt 0.2° / fov 59.5°
Venus crop for Golden bay
Venus1.9% rank 1pan 176.0° / tilt 21.1° / fov 25.9°
CACNet crop for Golden bay
CACNet3.8% rank 1pan 105.3° / tilt -29.8° / fov 58.3°
S53C59A29I19

Strongest: Composition. Weakest: Integrity. The composition is solid, with the subject placed horizontally centered and aligned well with the rule of thirds, ensuring a balanced and visually appealing arrangement. Despite the subject occupying only a small portion of the frame, its placement and alignment contribute significantly to the overall harmony of the image. However, the integrity score indicates that some elements might be partially cut off, which detracts slightly from the completeness of the scene.

PhotoGuider view preference heatmap for Golden bay
PhotoGuider
Venus view preference heatmap for Golden bay
Venus
CACNet view preference heatmap for Golden bay
CACNet
Rank 3

Neurathen rock castle

90.2% ranked PhotoGuider first 51 rankings

PhotoGuider crop for Neurathen rock castle
PhotoGuider90.2% rank 1pan -30.3° / tilt -0.2° / fov 59.5°
Venus crop for Neurathen rock castle
Venus5.9% rank 1pan 15.1° / tilt -30.0° / fov 60.0°
CACNet crop for Neurathen rock castle
CACNet3.9% rank 1pan 164.7° / tilt -26.9° / fov 57.2°
S54C37A67I3

Strongest: Aesthetics. Weakest: Integrity. The composition effectively balances the natural landscape with the man-made structure, creating a visually appealing contrast between the rugged cliffs and the serene forest. The subject, while not prominently featured, is still discernible and adds interest to the scene. However, the lack of complete visibility due to the dense foliage and the distance from the subject detracts from the overall integrity of the image.

PhotoGuider view preference heatmap for Neurathen rock castle
PhotoGuider
Venus view preference heatmap for Neurathen rock castle
Venus
CACNet view preference heatmap for Neurathen rock castle
CACNet
Rank 4

Dam wall

89.6% ranked PhotoGuider first 48 rankings

PhotoGuider crop for Dam wall
PhotoGuider89.6% rank 1pan -45.3° / tilt -0.2° / fov 59.5°
Venus crop for Dam wall
Venus8.3% rank 1pan -55.7° / tilt 25.6° / fov 53.6°
CACNet crop for Dam wall
CACNet2.1% rank 1pan 179.9° / tilt -13.4° / fov 57.9°
S57C84A73I24

Strongest: Composition. Weakest: Integrity. The composition excels with its balanced arrangement and adherence to the rule of thirds, where the dam structure is placed centrally but does not dominate the frame, allowing for a harmonious integration of the surrounding landscape. The subject's placement ensures it remains the focal point without being overwhelming, while the leading lines created by the road and the water flow guide the viewer's eye through the image, enhancing its visual appeal. Despite the subject occupying only a small portion of the frame, its strategic positioning within the thirds grid makes it stand out effectively.

PhotoGuider view preference heatmap for Dam wall
PhotoGuider
Venus view preference heatmap for Dam wall
Venus
CACNet view preference heatmap for Dam wall
CACNet
Rank 5

Preller drive

88.0% ranked PhotoGuider first 50 rankings

PhotoGuider crop for Preller drive
PhotoGuider88.0% rank 1pan -105.3° / tilt 15.2° / fov 59.5°
Venus crop for Preller drive
Venus4.0% rank 1pan -149.9° / tilt -30.0° / fov 60.0°
CACNet crop for Preller drive
CACNet8.0% rank 1pan -119.9° / tilt 5.6° / fov 58.3°
S59C62A41I25

Strongest: Composition and Subject. Weakest: Integrity. The composition is solid, with the subject placed horizontally centered and occupying approximately 5% of the frame. The subject is fully contained within the frame, ensuring no important elements are cut off. While the subject's prominence is solid, its small size within the frame might detract slightly from the overall aesthetic appeal.

PhotoGuider view preference heatmap for Preller drive
PhotoGuider
Venus view preference heatmap for Preller drive
Venus
CACNet view preference heatmap for Preller drive
CACNet
Rank 6

Konigsallee

87.8% ranked PhotoGuider first 49 rankings

PhotoGuider crop for Konigsallee
PhotoGuider87.8% rank 1pan 44.7° / tilt -0.2° / fov 59.5°
Venus crop for Konigsallee
Venus6.1% rank 1pan 30.1° / tilt -12.5° / fov 59.9°
CACNet crop for Konigsallee
CACNet6.1% rank 1pan 29.8° / tilt -13.3° / fov 58.4°
S69C70A67I21

Strongest: Composition and Subject. Weakest: Integrity. The composition excels with its balanced arrangement along the rule of thirds, creating a harmonious frame that guides the viewer's eye through the scene. The subject, while small, is prominently placed and fully contained within the frame, ensuring it remains the focal point. Despite the slight zoom-out, the subject's horizontal centering enhances the overall balance and symmetry of the image.

PhotoGuider view preference heatmap for Konigsallee
PhotoGuider
Venus view preference heatmap for Konigsallee
Venus
CACNet view preference heatmap for Konigsallee
CACNet
Rank 7

Shanghai bund

86.5% ranked PhotoGuider first 52 rankings

PhotoGuider crop for Shanghai bund
PhotoGuider86.5% rank 1pan -45.3° / tilt 0.2° / fov 59.5°
Venus crop for Shanghai bund
Venus3.8% rank 1pan 15.1° / tilt -30.0° / fov 60.0°
CACNet crop for Shanghai bund
CACNet9.6% rank 1pan 75.1° / tilt -9.1° / fov 59.7°
S65C39A48I32

Strongest: Subject. Weakest: Integrity. The subject, the cityscape, is prominently featured, drawing immediate attention due to its bright lights and distinctive architecture. The composition effectively uses the horizon line to separate the urban environment from the foreground, creating a sense of depth. However, the integrity score is low because the scene lacks some elements that might enhance its completeness, such as people interacting more dynamically or more varied architectural details closer to the viewer.

PhotoGuider view preference heatmap for Shanghai bund
PhotoGuider
Venus view preference heatmap for Shanghai bund
Venus
CACNet view preference heatmap for Shanghai bund
CACNet
Rank 8

Decor shop

85.7% ranked PhotoGuider first 49 rankings

PhotoGuider crop for Decor shop
PhotoGuider85.7% rank 1pan -120.0° / tilt -0.0° / fov 60.0°
Venus crop for Decor shop
Venus14.3% rank 1pan -37.2° / tilt -25.5° / fov 48.1°
CACNet crop for Decor shop
CACNet0.0% rank 1pan 45.2° / tilt -28.0° / fov 56.7°
S77C53A62I7

Strongest: Subject. Weakest: Integrity. The subject, which appears to be a workspace, is prominently featured in the image, drawing attention to its details and contents. The composition is balanced, with the desk and chair positioned centrally, creating a sense of order amidst the clutter. However, the integrity score is low due to the lack of complete visibility of some elements, such as the wall art and the bookshelves, which are partially obscured by the framing.

PhotoGuider view preference heatmap for Decor shop
PhotoGuider
Venus view preference heatmap for Decor shop
Venus
CACNet view preference heatmap for Decor shop
CACNet
Rank 9

Schadowplatz

85.4% ranked PhotoGuider first 48 rankings

PhotoGuider crop for Schadowplatz
PhotoGuider85.4% rank 1pan 30.3° / tilt 14.8° / fov 59.5°
Venus crop for Schadowplatz
Venus14.6% rank 1pan 0.1° / tilt 2.6° / fov 59.9°
CACNet crop for Schadowplatz
CACNet0.0% rank 1pan -150.0° / tilt 21.6° / fov 58.0°
S64C32A77I5

Strongest: Aesthetics. Weakest: Integrity. The image's strong aesthetic appeal comes from its balanced framing and the harmonious integration of architectural elements with the surrounding environment. The subject, a clothing store, is prominently displayed through large windows that showcase its interior, drawing attention to the merchandise inside. However, the integrity score is low due to the lack of complete scene elements, which suggests that some parts of the building or adjacent areas might be missing from the frame, affecting the overall completeness of the scene.

PhotoGuider view preference heatmap for Schadowplatz
PhotoGuider
Venus view preference heatmap for Schadowplatz
Venus
CACNet view preference heatmap for Schadowplatz
CACNet
Rank 10

Signal hill sunrise

85.1% ranked PhotoGuider first 47 rankings

PhotoGuider crop for Signal hill sunrise
PhotoGuider85.1% rank 1pan 165.0° / tilt -0.0° / fov 60.0°
Venus crop for Signal hill sunrise
Venus14.9% rank 1pan 30.1° / tilt 9.8° / fov 59.7°
CACNet crop for Signal hill sunrise
CACNet0.0% rank 1pan -75.0° / tilt -33.0° / fov 57.8°
S63C73A84I12

Strongest: Aesthetics. Weakest: Integrity. The composition effectively uses the rule of thirds, placing the car slightly off-center but still within the frame's boundaries. The subject, a white car, is positioned to draw attention while maintaining a balanced view of the mountainous backdrop. Despite the weak integrity score, the subject's placement ensures it remains prominent without being cut off, contributing to the overall visual appeal.

PhotoGuider view preference heatmap for Signal hill sunrise
PhotoGuider
Venus view preference heatmap for Signal hill sunrise
Venus
CACNet view preference heatmap for Signal hill sunrise
CACNet

Remaining 90 scenes

Explore the full study

Showing 90 scenes

PhotoGuider
Venus
CACNet

PhotoGuider
Venus
CACNet